Copilot CLI

AI coding assistants store complete conversation history and operation records locally in plaintext (JSON or plain text) without encryption.

Unspill is a Go language CLI tool for scanning the local history records of AI coding assistants such as Copilot CLI, Claude Code, and Codex to find leaked credentials.

To enable named display, organizations need to ensure that the Copilot CLI version is at least 1.0.97, or the Copilot SDK version is at least 1.0.21.

After entering the agent command in Copilot CLI and selecting a scope (project-level, user-level, or organization-level), Copilot automatically completes the scaffolding for the custom Agent, including writing the description, formatting the front matter, and structuring success criteria and methodology.

Copilot CLI supports a headless mode, passing a one-shot prompt via -p, which can be placed in a GitHub Actions runner or any CI system and triggered based on repository events.

The explorer agent built into Copilot CLI uses Haiku-class small models to explore the codebase, avoiding the use of large models for low-value tasks.

Copilot CLI supports the slash remote command, allowing users to continue controlling the same session from their phone after closing their laptop.

The rubber duck agent built into Copilot CLI uses GPT to review plans written by Opus, leveraging differences in the training data of different models for cross-validation.

Browy's architecture comprises three programs: a browser extension, a Node host, and the Copilot CLI, which cooperate through a native messaging channel and JSON-RPC.

The copilot CLI's built-in tools such as bash and web_fetch are kept off the allowlist by default in Browy and need to be manually enabled.

For Codex and Copilot CLI outside Claude Code, whether shim-cli equally implements "masking before reading" or only operates at the observation level requires developers to verify themselves.

shim-cli states that it supports Claude Code, Codex, Copilot CLI and VS Code.

When handling sensitive code in regulated environments, Copilot CLI's default sandbox is a more prudent starting point; Claude Code's permission model is more transparent and suits scenarios that pursue maximum flexibility and can accept case-by-case confirmation; Codex lies between the two.

Copilot CLI's sandbox is more explicit about network control, with outbound access constrained by policy, tending to keep sensitive information outside the sandbox.

Copilot CLI by default provides a layer of isolation on supported platforms, confining AI-initiated commands to a controlled scope.

Copilot CLI supports a per-task cost cap via --max-ai-credits, and CLI 1.0.89 rejects any value below 30 credits.

HydraFusion is only available through three entry points—Copilot CLI, VS Code 1.140 and above, and the GitHub Copilot app—and does not include cloud agents, JetBrains IDE, Visual Studio, or Copilot Chat on github.com.

CLI 1.0.89 emits events such as session.fusion_resolved and assistant.fusion_phase_started when using --output-format json, but this schema is not documented.

Watchbird targets mainstream CLI-based AI coding assistants, with official support explicitly mentioned for Claude Code, Codex, Copilot CLI, and others, covering more than ten tools in total

The WorkIQ developer tool is essentially a wrapper around the Agents Toolkit and is modeled as a plugin for the Copilot CLI.